Memory Benchmarks

This page defines how smarna evaluates memory quality for long-running AI agents and MCP server environments.

Metrics

• Time-to-degradation under sustained interactions
• Token efficiency vs. baseline retrieval approaches
• Recall accuracy for old and recent facts
• Latency stability across long horizons

Evaluation setup

We compare memory quality across equivalent workloads with controlled token budgets and fixed evaluation prompts.

Back to homepage